DeepSeek AI, China’s Tech Advancement, NVIDIA Stock Drop: Is This the End of GPU Dominance?

"It’s like a meteorite had hit Earth" - The DeepSeek AI Revolution

Feb 04, 2025

In this episode, the hosts discuss the recent launch of DeepSeek, a groundbreaking AI model with significant implications for the tech industry and geopolitical landscape. They explore its technical innovations, the market’s reaction, and the potential for smaller players to compete in AI development.

The conversation also delves into the importance of trust in AI models, the future of AI innovation, and insights for entrepreneurs navigating this rapidly evolving space.

Chapters

00:00 Introduction and Overview of DeepSeek
02:01 DeepSeek’s Technical Innovations and Implications
05:48 Geopolitical Context and Market Reactions
09:51 Adoption and Trust in AI Models
13:55 The Future of AI Development and Innovation
18:06 Entrepreneurial Insights and Market Dynamics

Takeaways

DeepSeek’s Breakthrough Challenges AI Development Norms
DeepSeek R1 marks a significant shift in AI by eliminating human involvement in training through a teacher-student model. This challenges traditional reinforcement learning with human feedback (RLHF) and showcases that high-quality AI can be developed with minimal manual oversight.
Geopolitical and Market Disruptions
The launch of DeepSeek has had major geopolitical and economic implications, causing a drop in NVIDIA’s stock and raising concerns in the West about China’s growing AI capabilities. The open-source nature of DeepSeek also introduces new dynamics in AI accessibility and competition.
Trust and Cost Efficiency Shape AI Adoption
While DeepSeek offers powerful capabilities at a fraction of the cost of Western models, enterprise adoption depends on trust, security, and regulatory considerations. Open-source AI democratizes access, but businesses must balance innovation with compliance and practical implementation.

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction & Welcome

Maxson Tee: Welcome back to another episode of the CRM show, the Chris Rod Max show. We have my co-host with me today, Chris Rod. Welcome back.

Chris: Hey guys! Hi!

Rod: Hi everyone.

Setting the Stage: Recent Developments

Maxson Tee: This week has been packed with significant developments. Following last week’s inauguration, we witnessed the launch of DeepSeek, which we’ll discuss in detail today. Interestingly, this led to some controversy in the West, with OpenAI claiming DeepSeek trained its model using OpenAI’s model. We’ve also seen interesting discussions about ChatGPT rappers, which were previously criticized but are now gaining acceptance. Let’s dive into these topics.

The biggest story this week is definitely DeepSeek. It’s notable that it caused a 600 million or billion drop in NVIDIA’s value. Rod, I see you’re ready to comment - what makes DeepSeek so special?

DeepSeek’s Revolutionary Impact

Rod: Max, while you’re calling this the news of the week, I’d go further and say this could be the news of the year. The impact is comparable to a meteorite hitting Earth - that’s how significant this is.

Maxson Tee: That’s a big claim, considering we’re just at the start of the year.

Rod: Indeed, it’s been an eventful year, but I believe when we look back at the end of 2025, DeepSeek R1 will stand among the most relevant and impactful developments. While there’s a geopolitical aspect we could discuss, I’ll focus on the technical significance: DeepSeek has demonstrated, for the first time, that you don’t need human involvement to create better models. Historically, with OpenAI models, human feedback was essential - indicating good or bad answers, guiding the model’s direction. DeepSeek R1 achieved its results through a teacher-student model approach, eliminating the need for human review and assessment.

Non-Technical Implications

Maxson Tee: Interesting. Chris, from a non-technical perspective, what are your thoughts on DeepSeek’s implications?

Chris: As someone who does a lot of hiking and trail running, I’ve been listening to numerous podcasts, and this is undoubtedly the top news right now. What’s fascinating is how people are trying to dissect DeepSeek, searching for potential flaws or questioning its legitimacy. The remarkable fact is that it’s legitimate and achieved at a cost of $6 billion versus roughly $100 billion for comparable models.

I think the market’s reaction reveals significant fear, especially given the current geopolitical dynamics between the West and China. While many are surprised that China has developed a model matching Western capabilities, this development isn’t entirely unexpected. The real question now lies in adoption and future user implementation.

Market Impact and Model Commoditization

Maxson Tee: Absolutely agree, Chris. We’re witnessing a commoditization of AI models, with Alibaba and others releasing their own versions claiming superiority. The key factor becoming increasingly important is trust in these models and their adoption over time.

Referencing Nassim Taleb’s recent Bluebook interview, he pointed out that while many innovators create initial technologies, the early innovators often don’t capture most of the value. Instead, value comes from adoption and building valuable applications on top of the technology.

Rod, from your technical perspective, given your experience in building AI models, how do you view the balance between trust issues and cost implications?

Technical Implementation and Accessibility

Rod: We need to distinguish between two aspects. First, there’s the web-based chatbot similar to ChatGPT, which raises data hosting concerns since it’s based in China/Hong Kong rather than the US or Europe. This might limit its use for certain industries or companies.

However, it’s also an open-weights model, meaning anyone can download and run it locally without sharing data externally. Remarkably, estimates suggest you can build a $6,000 gaming-level computer capable of running their most powerful model at full capacity. While it might be slightly sluggish at eight tokens per second, it’s still productive - a dramatic shift from requiring billion-dollar infrastructure.

Maxson Tee: That’s fascinating - it’s similar to downloading Bitcoin’s code to run your own mining system. Now you can do the same with AI models at home.

China’s Approach to AI Development

Chris: This connects to our earlier discussions about China’s approach to AI. It’s remarkable that DeepSeek’s model is open source, allowing anyone to download and contribute to its development. We previously discussed how China builds AI solutions vertically by industry, fostering collaboration on datasets and training within specific sectors like shipping or banking. Now we’re seeing these approaches converge.

Innovation and Global Competition

Maxson Tee: There’s an interesting irony here - China, typically seen as more centralized, is taking an open-source approach, while the West, traditionally more distributed, has concentrated AI development among a few players. DeepSeek, coming from a hedge fund rather than a major tech company, demonstrates interesting potential.

This shows that smaller players can compete effectively - it’s not just about money and computing power. Countries like the UK and France, with sufficient intellectual capital, could potentially develop similar models. This creates more opportunities for innovation across different countries and better pricing from an economic perspective.

Enterprise Adoption and Implementation

Rod: Max, given your perspective on both enterprise and startup sides, how are companies approaching implementation? Are they actively exploring ways to benefit from it, or are they hesitant due to its foreign origin?

Maxson Tee: From a model perspective, I’m confident many AI departments are examining DeepSeek due to its open-source nature. In financial services, many technologies are derived from open-source software, including risk software. When an AI model is fully open source and trustworthy, adoption typically follows.

Regarding use cases, banks are showing flexibility about which models they use - they care more about results and cost efficiency than the specific model. This cost aspect is crucial for growth, as lower costs enable more innovation attempts and higher chances of breakthrough successes.

Market Response and Technical Innovation

Chris: The impact of DeepSeek varies by audience. We saw significant stock market drops in technology and semiconductor sectors, partly because DeepSeek doesn’t require the latest GPU chips. However, for enterprises, the specific model matters less than its effectiveness.

From a research perspective, it’s interesting how export restrictions may have pushed China to develop more creative, cost-effective solutions. notably, DeepSeek seems to have overcome data limitations through innovative self-learning algorithms.

Technical Details and Market Implications

Rod: Regarding the teacher-student model approach, there are rumors that DeepSeek might have used OpenAI’s models as teachers, though this isn’t confirmed and would violate OpenAI’s terms of service. Some users report the model occasionally identifying itself as an OpenAI model, suggesting possible training connections.

Regarding market impact, Nvidia’s stock remains down about 15% since this news. This raises questions about future GPU requirements and the entire AI development model, including initiatives like Stargate’s $500 billion investment.

Future Implications and Applications

Maxson Tee: The requirements really depend on the goal. Building AGI might need massive computation, but solving daily problems often doesn’t require top-end capabilities. The market reaction follows typical patterns - initial excitement, potential exuberance, then adjustment.

Regarding applications, Greg Eisenberg’s observation about ChatGPT wrappers is interesting - what was initially criticized as simple repackaging now appears to be an effective approach, providing direct solutions to specific customer needs with commoditized models.

Closing Thoughts on Implementation

Chris: It’s ultimately about the interface - creating something people want to integrate into their daily lives. The goal is to own real estate on people’s phones, becoming their go-to solution, regardless of the underlying engine.

Maxson Tee: Exactly - like transportation, users care about getting from A to B, not the specific vehicle.

Rod: From an enterprise perspective, the real challenge is implementation - ensuring frontline workers can use this technology effectively. While AI might threaten certain consulting roles, the moat lies in relationships and system integration. The advantage comes from providing this connection and maintaining customer relationships.

Maxson Tee: Indeed - “whoever controls your distribution controls your life.” This presents opportunities to rethink software deployment, with AI potentially helping bridge legacy systems. We’ll continue this discussion next week. Remember to like and subscribe, and share your thoughts about DeepSeek in the comments.

Your Hosts

Christine Wang Rod Rivera Maxson J.Y. Tee

Follow Chris Rod Max:

https://linkedin.com/in/christinewang0/

https://linkedin.com/in/rodriveracom/

https://linkedin.com/in/maxsontjy/

Follow on X:

https://x.com/chrisrodmax

Follow on Instagram:

https://instagram.com/chrisrodmax

Follow on LinkedIn:

https://linkedin.com/company/chrisrodmax